No policy ensemble or domain set. Skipping action prediction and execution

Rasa Open Source
22

image
image1739×58 4.71 KB

I get the following error, can you help me?

23

Hi @NikoAP77, welcome to the forum!

When do you receive this error, what is the output of rasa --version, and have you tried fixing the problem with the solution posted here?

24

Mi version es: 2.0.2

Exacto ya seguí los pasos de la solución pero me sigue dando el error.

25

My Spanish is a bit rusty, but I think you’re saying that you’ve followed the steps (ran rasa train) and still get that error on rasa shell?

In that case, can you please post (a) the output of rasa data validate, and (b) the output of ls ./models?

26

Hi everyone,

I’m facing the same problem migrating from Rasa-X 0.37.0 to 0.41.0 and rasa fom 2.3.1 to 2.7.1

My model uses Spacy, i run in local “rasa data validate -vv” and “rasa train -vv” and all looks good a part that I went to see inside the archived model. The Response selector*.json and the TED*.json files are empty and there are no spacy-linked folders.

The config of my model is like this :

language: fr
pipeline:
  - case_sensitive: false
    model: fr_core_news_md
    name: SpacyNLP
  - intent_split_symbol: +
    intent_tokenization_flag: false
    name: SpacyTokenizer
  - dimensions:
      - time
      - number
      - amount-of-money
      - ordinal
      - distance
      - duration
      - email
      - phone-number
    locale: fr_FR
    name: DucklingEntityExtractor
    timeout: 15
    timezone: “Europe/Paris”
    url: 'http://customized-rasa-x-duckling:8000'
  - case_sensitive: false
    name: RegexFeaturizer
    use_lookup_tables: true
    use_regexes: true
  - case_sensitive: false
    name: RegexEntityExtractor
    use_lookup_tables: true
    use_regexes: true
  - name: SpacyFeaturizer
    pooling: max
  - dimensions:
      - PERSON
      - LOC
      - ORG
      - PRODUCT
    name: SpacyEntityExtractor
  - name: EntitySynonymMapper
  - features:
      - - low
        - title
        - upper
      - - BOS
        - EOS
        - low
        - upper
        - title
        - digit
      - - low
        - title
        - upper
    name: LexicalSyntacticFeaturizer
  - analyzer: char_wb
    lowercase: true
    max_ngram: 4
    min_ngram: 1
    name: CountVectorsFeaturizer
    use_shared_vocab: false
  - name: SklearnIntentClassifier
    C:
      - 1
      - 2
      - 5
      - 10
      - 20
      - 100
    kernels:
      - linear
    gamma:
      - 0.1
    max_cross_validation_folds: 5
    scoring_function: f1_weighted
  - name: DIETClassifier
    epochs: 100
    constrain_similarities: true
    intent_classification: true
    entity_recognition: true
    model_confidence: linear_norm
    connection_density: 0.2
  - name: FallbackClassifier
    threshold: 0.6
    ambiguity_threshold: 0.02
  - name: ResponseSelector
    retrieval_intent: faq
    epochs: 100
    constrain_similarities: true
    connection_density: 0.2
    model_confidence: linear_norm
policies:
  - name: RulePolicy
    core_fallback_threshold: 0.3
    core_fallback_action_name: action_default_fallback
    enable_fallback_prediction: true
    restrict_rules: true
    check_for_contradictions: true
  - name: MemoizationPolicy
    max_history: 6
    priority: 3
  - name: TEDPolicy
    max_history: 6
    epochs: 200
    constrain_similarities: true
    connection_density: 0.2
    model_confidence: linear_norm

I also renamed the call to spacy inside the dockerfile which creates our personnal rasa-spacy-fr image so that it does not link the spacy pre-trained to “FR”

Acually, rasa x achieve to train a model with that config, but when i talk to the bot, it only catches not implemented intents with the same name as the user utterance and no action is taken in the wake, for instance the user says “Hey” and that’s how the production pod manages that : Published Pika events to exchange ‘rasa-exchange’ on host ‘customized-rabbit’: {‘sender_id’: ‘f86ca803311f4e7595358256a905eabe’, ‘event’: ‘user’, ‘timestamp’: 1626085054.9291916, ‘text’: ‘hey’, ‘parse_data’: {‘intent’: {‘name’: ‘hey’, ‘confidence’: 1.0}, ‘entities’: [], ‘text’: ‘/hey’, ‘message_id’: ‘57d5508079274ce3a92b65f11c57d847’, ‘metadata’: {}, ‘intent_ranking’: [{‘name’: ‘hey’, ‘confidence’: 1.0}]}, ‘input_channel’: ‘rasa’, ‘message_id’: ‘57d5508079274ce3a92b65f11c57d847’, ‘metadata’: {}}

(while it might find a greet intent and start a meeting form)

I tried to talk to the bot with the shell command too and it seems that it has too great fallback scores, even with those prediction, the model does not return a fallback intent but founds intents I never implemented…

Do someone know what could be the reason of this behaviour ?